多智能体辩论用于可解释交易：推理、共识与模拟市场表现
原标题：Multi-Agent Debate for Explainable Trading: Reasoning, Consensus, and Performance in Simulated Markets
AI 摘要
研究团队构建了一个多智能体辩论框架，用于历史市场模拟中的投资组合配置，让专门化智能体在结构化推理与干预协议下提出、批评并修订投资决策。在210次受控实验中，聚合推理质量与夏普比率和总回报均无显著关系，结构化提示虽大幅提升推理质量，但收益并未稳定提高。研究识别出「谄媚式收敛」这一核心失败模式，并发现保留分歧的JSD干预能提升夏普比率和索提诺比率，而强制因果推理的干预无效。结论认为多智能体辩论的价值更多来自保留独立信息信号，而非提升个体推理质量。
正文节选
Multi-Agent Debate for Explainable Trading : Reasoning, Consensus, Performance in Simulated Markets Abstract Large language models (LLMs) are increasingly used for financial decision-making, yet it remains unclear whether improvements in their reasoning quality translate into better economic outcomes. We investigate this question through a multi-agent debate framework for portfolio allocation in historical market simulations, where specialized agents propose, critique, and revise investment deci